Is the AI race already over? Well, not really. However, while we have all been focused on the big-name companies in Silicon Valley, China is methodically improving its AI capabilities while not putting it behind any paywalls. Moonshot AI, a Beijing-based company, has developed Kimi-Researcher, an autonomous agent that thinks through a problem the way a junior analyst would.
What is Kimi-Researcher by Moonshot AI?
Moonshot AI's Kimi-Researcher is an autonomous AI agent with impressive abilities in handling complex research tasks. This AI agent approaches problems like a real researcher, and its performance on difficult benchmarks is turning heads.
Kimi‑Researcher scored 26.9% Pass@1 score on Humanity's Last Exam (HLE), edging past Google Gemini Deep Research and OpenAI's o3 with search tools. HLE is widely regarded as the most challenging public test of long-form reasoning, drawing questions from hundreds of academic and professional fields.
Kimi hit that mark by running dozens of self-directed reasoning loops, scanning hundreds of primary sources, and rewriting its plan until every contradiction was resolved, all while remaining completely free to use.
How did Kimi-Researcher achieve that?
Moonshot trained the Kimi research agent end‑to‑end with agentic reinforcement learning on top of its in‑house Kimi language model. Instead of memorizing static prompt chains, the system was rewarded for factual accuracy, coherent arguments, and well-structured citations, making the AI agent feel less like a chatbot and more like a meticulous desk researcher.
Here are some of its key features and functions:
- Deep Dive Research: According to an AIBase article, for a single task, Kimi-Researcher can perform an average of 23 reasoning steps and research over 200 URLs to collect and synthesize information.
- Advanced Training Method: The agent is trained entirely through end-to-end reinforcement learning, which helps it to adapt and refine its problem-solving strategies on its own.
- Benchmark Performance: It has shown massive improvement on complex benchmarks, such as "Humanity's Last Exam," scoring 26.9% compared to its initial score of 8.6%, and 69% on xbench-DeepSearch, topping OpenAI, Gemini, and DeepSeek.
- Practical Applications: It is made for real-world research tasks such as performing due diligence, learning new subjects from the ground up, and conducting academic analysis.
How to use Kimi-Researcher for free:
Getting started with Kimi-Researcher is simpler than its technical background. Go to the Kimi AI website and sign up using your Google account.
Once you have done that, click on the 🔬 Researcher option within the chatbox and enter your research question. For the purpose of this article, we will keep it simple and ask What is an AI research agent?
It took Kimi a solid few minutes to create the full report, but the end result was a well-structured and cited report.
Conclusion:
Most companies have started making deep research a free option, but it always felt like a less better version of deep research than the paid version. Kimi-Researcher has changed the narrative, and for the first time, an AI deep research agent felt as capable and powerful and on par with those locked behind paywalls. The takeaway for users is simple: serious, multi‑step AI research no longer costs $20 a month, and it is now free for anyone to experience. Moonshot AI demonstrates that autonomous reasoning is no longer a boutique feature; it's a baseline capability.
